Thu. 07/08 – GPT-5 Released
GPT-5 launch: what changed with OpenAI's unified reasoning model
OpenAI unveiled GPT-5, positioned as a unified model combining fast responses with deeper reasoning. Sam Altman framed GPT-5 as a step toward broadly capable, near-AGI systems — describing interactions as feeling like a PhD-level expert. The release consolidates prior model families into one experience, using an internal router that automatically switches to a reasoning configuration for complex, multi-step tasks. That router aims to simplify user experience while enabling agent-like behaviors such as generating software applications and completing workflows on users’ behalf.
Developer tiers, pricing, and API control
OpenAI announced three API sizes (GPT-5, GPT-5 mini, GPT-5 nano) and new verbosity controls for response length. Token pricing was disclosed: roughly $1.25 per million input tokens and $10 per million output tokens for the base model. Free users can access GPT-5 with a usage cap, while paid subscribers get higher limits and GPT-5 Pro options for more compute and reliability. These changes affect cost modeling for startups, enterprises, and AI-driven products.
Safer outputs, fewer hallucinations, and 'safe completions'
OpenAI reports materially improved safety metrics: hallucination rates dropped substantially on benchmark tests, and GPT-5 introduces “safe completions” that aim to provide helpful but non-actionable answers when queries could be harmful. The model also more often admits uncertainty. While the model still hallucinates and can be deceptive in edge cases, OpenAI emphasized extensive safety testing and lower deception rates than previous models.
Software on demand and coding benchmarks
GPT-5 scored highly on coding benchmarks, outperforming competitors on some tests and demonstrating rapid code generation in demos. OpenAI positions the model as enabling “software on demand,” where the model can spin up working applications and handle multi-step engineering tasks—potentially reducing development time and changing workflows in product teams and agencies.
Politics and manufacturing: tariffs, Apple, and Intel scrutiny
On the policy side, President Trump proposed roughly 100% tariffs on chips and semiconductors, while offering exemptions to companies that commit to U.S. manufacturing. Apple publicly pledged another $100 billion for domestic projects and a partnership with Corning to make 100% of iPhone and Apple Watch cover glass in the U.S., a notable reshoring step. Separately, Intel CEO Pat Gelsinger’s successor (Lip-Bu Tan) faced calls for resignation after scrutiny of past China ties, adding political pressure to an already sensitive chip sector.
Why this episode matters
- Developers and product managers must reassess cost, latency, and safety trade-offs with GPT-5 APIs.
- Enterprise leaders should test GPT-5 Pro for mission-critical workflows and compliance monitoring.
- Supply-chain and policy teams need to track tariff exemptions tied to U.S. manufacturing commitments.
In short, this episode captures a pivotal day where AI capabilities, commercial strategy, and geopolitical policy intersect—shaping product roadmaps, regulatory risk, and where chips and glass for phones get made.
